Experimental investigation and prediction of longterm behavior of textile reinforced concrete for strengthening: Fid-Bau Portal

Experimental investigation and prediction of longterm behavior of textile reinforced concrete for strengthening

Freitag, S. / Beer, M. / Jesse, F. / Weiland, S.

Textile reinforced concrete (TRC) represents a proper material for retrofitting and strengthening of aged or damaged reinforced concrete structures. The practical application of TRC strengthening presupposes a numerical prediction of the improved structural behavior including the material behavior of the TRC strengthening layer. Data and numerical models for predicting the long-term behavior of TRC, however, are not available to a sufficient amount and reliability. Long-term tests with small TRC specimens under uniaxial tension were investigated under sustained loading for up to six months. The test results provide some insight into long-term performance of TRC. But experiments covering time periods with relevance to structural life-time, however, are not realizable. First creep tests of TRC specimens under uniaxial tension have been carried out. The resulting strain-time curves clearly show a slowing strain rate during the primary and secondary strain increase phases. However, there is a lack of information about the mechanisms behind, while obviously several mechanisms contribute simultaneously, interact with each other but cannot be separated based on the experimental observations. Because of restrictions in time and in resources only a fraction of the data required for a traditional reliable numerical modeling of the long-term behavior can be gained experimentally. Thus, a combination of specially selected experiments and an innovative numerical method for evaluating rare data over only short-time periods is pursued to solve the problem. For predicting long-term behavior of textile reinforced concrete directly from experimental data neural networks represent a viable approach. They are particularly powerful in cases in which physical material properties cannot be identified or specified properly in short-term tests. Neural networks operate without a model specification. Further developments of the presented procedure focus on extensions for applicability to nonstationary data series based on a network based detrending method.